signal: Use GROUP_STOP_PENDING to stop once for a single group stop
authorTejun Heo <tj@kernel.org>
Wed, 23 Mar 2011 09:37:00 +0000 (10:37 +0100)
committerTejun Heo <tj@kernel.org>
Wed, 23 Mar 2011 09:37:00 +0000 (10:37 +0100)
Currently task->signal->group_stop_count is used to decide whether to
stop for group stop.  However, if there is a task in the group which
is taking a long time to stop, other tasks which are continued by
ptrace would repeatedly stop for the same group stop until the group
stop is complete.

Conversely, if a ptraced task is in TASK_TRACED state, the debugger
won't get notified of group stops which is inconsistent compared to
the ptraced task in any other state.

This patch introduces GROUP_STOP_PENDING which tracks whether a task
is yet to stop for the group stop in progress.  The flag is set when a
group stop starts and cleared when the task stops the first time for
the group stop, and consulted whenever whether the task should
participate in a group stop needs to be determined.  Note that now
tasks in TASK_TRACED also participate in group stop.

This results in the following behavior changes.

* For a single group stop, a ptracer would see at most one stop
  reported.

* A ptracee in TASK_TRACED now also participates in group stop and the
  tracer would get the notification.  However, as a ptraced task could
  be in TASK_STOPPED state or any ptrace trap could consume group
  stop, the notification may still be missing.  These will be
  addressed with further patches.

* A ptracee may start a group stop while one is still in progress if
  the tracer let it continue with stop signal delivery.  Group stop
  code handles this correctly.

Oleg:

* Spotted that a task might skip signal check even when its
  GROUP_STOP_PENDING is set.  Fixed by updating
  recalc_sigpending_tsk() to check GROUP_STOP_PENDING instead of
  group_stop_count.

* Pointed out that task->group_stop should be cleared whenever
  task->signal->group_stop_count is cleared.  Fixed accordingly.

* Pointed out the behavior inconsistency between TASK_TRACED and
  RUNNING and the last behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
fs/exec.c
include/linux/sched.h
kernel/signal.c

index 5e62d26a4fecec227d81700b0b9fd6542b715ad6..8328beb9016f909dc9cef5b169fb8fc33a7fcb8f 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -1659,6 +1659,7 @@ static int zap_process(struct task_struct *start, int exit_code)
 
        t = start;
        do {
+               task_clear_group_stop_pending(t);
                if (t != current && t->mm) {
                        sigaddset(&t->pending.signal, SIGKILL);
                        signal_wake_up(t, 1);
index 85f51042c2b8f52e41add5a4c941a0099b3ea618..b2a17dfbdbad5da0af6866045cfefd6b029d8294 100644 (file)
@@ -1775,8 +1775,11 @@ extern void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *
 /*
  * task->group_stop flags
  */
+#define GROUP_STOP_PENDING     (1 << 16) /* task should stop for group stop */
 #define GROUP_STOP_CONSUME     (1 << 17) /* consume group stop count */
 
+extern void task_clear_group_stop_pending(struct task_struct *task);
+
 #ifdef CONFIG_PREEMPT_RCU
 
 #define RCU_READ_UNLOCK_BLOCKED (1 << 0) /* blocked while in RCU read-side. */
index ecb20089eaffb8baef7e7add27a25ccb0cec5384..a2e7a6527d2471abae2e1f6f1c5cc9495d7d86ff 100644 (file)
@@ -124,7 +124,7 @@ static inline int has_pending_signals(sigset_t *signal, sigset_t *blocked)
 
 static int recalc_sigpending_tsk(struct task_struct *t)
 {
-       if (t->signal->group_stop_count > 0 ||
+       if ((t->group_stop & GROUP_STOP_PENDING) ||
            PENDING(&t->pending, &t->blocked) ||
            PENDING(&t->signal->shared_pending, &t->blocked)) {
                set_tsk_thread_flag(t, TIF_SIGPENDING);
@@ -232,19 +232,19 @@ static inline void print_dropped_signal(int sig)
  * CONTEXT:
  * Must be called with @task->sighand->siglock held.
  */
-static void task_clear_group_stop_pending(struct task_struct *task)
+void task_clear_group_stop_pending(struct task_struct *task)
 {
-       task->group_stop &= ~GROUP_STOP_CONSUME;
+       task->group_stop &= ~(GROUP_STOP_PENDING | GROUP_STOP_CONSUME);
 }
 
 /**
  * task_participate_group_stop - participate in a group stop
  * @task: task participating in a group stop
  *
- * @task is participating in a group stop.  Group stop states are cleared
- * and the group stop count is consumed if %GROUP_STOP_CONSUME was set.  If
- * the consumption completes the group stop, the appropriate %SIGNAL_*
- * flags are set.
+ * @task has GROUP_STOP_PENDING set and is participating in a group stop.
+ * Group stop states are cleared and the group stop count is consumed if
+ * %GROUP_STOP_CONSUME was set.  If the consumption completes the group
+ * stop, the appropriate %SIGNAL_* flags are set.
  *
  * CONTEXT:
  * Must be called with @task->sighand->siglock held.
@@ -254,6 +254,8 @@ static bool task_participate_group_stop(struct task_struct *task)
        struct signal_struct *sig = task->signal;
        bool consume = task->group_stop & GROUP_STOP_CONSUME;
 
+       WARN_ON_ONCE(!(task->group_stop & GROUP_STOP_PENDING));
+
        task_clear_group_stop_pending(task);
 
        if (!consume)
@@ -765,6 +767,9 @@ static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
                t = p;
                do {
                        unsigned int state;
+
+                       task_clear_group_stop_pending(t);
+
                        rm_from_queue(SIG_KERNEL_STOP_MASK, &t->pending);
                        /*
                         * If there is a handler for SIGCONT, we must make
@@ -906,6 +911,7 @@ static void complete_signal(int sig, struct task_struct *p, int group)
                        signal->group_stop_count = 0;
                        t = p;
                        do {
+                               task_clear_group_stop_pending(t);
                                sigaddset(&t->pending.signal, SIGKILL);
                                signal_wake_up(t, 1);
                        } while_each_thread(p, t);
@@ -1139,6 +1145,7 @@ int zap_other_threads(struct task_struct *p)
        p->signal->group_stop_count = 0;
 
        while_each_thread(p, t) {
+               task_clear_group_stop_pending(t);
                count++;
 
                /* Don't bother with already dead threads */
@@ -1690,7 +1697,7 @@ static void ptrace_stop(int exit_code, int why, int clear_code, siginfo_t *info)
         * If there is a group stop in progress,
         * we must participate in the bookkeeping.
         */
-       if (current->signal->group_stop_count > 0)
+       if (current->group_stop & GROUP_STOP_PENDING)
                task_participate_group_stop(current);
 
        current->last_siginfo = info;
@@ -1775,8 +1782,8 @@ static int do_signal_stop(int signr)
        struct signal_struct *sig = current->signal;
        int notify = 0;
 
-       if (!sig->group_stop_count) {
-               unsigned int gstop = GROUP_STOP_CONSUME;
+       if (!(current->group_stop & GROUP_STOP_PENDING)) {
+               unsigned int gstop = GROUP_STOP_PENDING | GROUP_STOP_CONSUME;
                struct task_struct *t;
 
                if (!likely(sig->flags & SIGNAL_STOP_DEQUEUED) ||
@@ -1796,8 +1803,7 @@ static int do_signal_stop(int signr)
                         * stop is always done with the siglock held,
                         * so this check has no races.
                         */
-                       if (!(t->flags & PF_EXITING) &&
-                           !task_is_stopped_or_traced(t)) {
+                       if (!(t->flags & PF_EXITING) && !task_is_stopped(t)) {
                                t->group_stop = gstop;
                                sig->group_stop_count++;
                                signal_wake_up(t, 0);
@@ -1926,8 +1932,8 @@ relock:
                if (unlikely(signr != 0))
                        ka = return_ka;
                else {
-                       if (unlikely(signal->group_stop_count > 0) &&
-                           do_signal_stop(0))
+                       if (unlikely(current->group_stop &
+                                    GROUP_STOP_PENDING) && do_signal_stop(0))
                                goto relock;
 
                        signr = dequeue_signal(current, &current->blocked,
@@ -2073,7 +2079,7 @@ void exit_signals(struct task_struct *tsk)
                if (!signal_pending(t) && !(t->flags & PF_EXITING))
                        recalc_sigpending_and_wake(t);
 
-       if (unlikely(tsk->signal->group_stop_count) &&
+       if (unlikely(tsk->group_stop & GROUP_STOP_PENDING) &&
            task_participate_group_stop(tsk))
                group_stop = CLD_STOPPED;
 out: